应用遗传算法构建化学模式分类器  

The application of genetic algorithm to the construction of chemical pattern classifier

作　　者：廖兴发[1] 陈德钊[1] 贺益君[1] 

机构地区：[1]浙江大学化学工程系仿真中心,浙江杭州310027

出　　处：《计算机与应用化学》2007年第5期593-596,共4页Computers and Applied Chemistry

基　　金：国家自然科学基金资助(20276063)

摘　　要：神经网络和统计分析所构建的分类器均为复杂算式,难以体现专业知识;而分类规则直接以属性值为条件,确定个体类别,易于专业分析。对于连续属性的样本数据,本文应用基于信息熵的Chi-merge方法将其离散化,并将提取最优规则转换为组合优化问题,进而采用遗传算法求解。为此,本文将规则提取演绎为种群进化,并设计了个体适应度函数。由此提取出最优的分类规则,经过修剪处理后,与判别准则一起构成模式分类器。本文将其应用于橄榄油产地判别,所建立的分类器简单明了,规则数少,性能良好,适用于化学模式分类。Classifiers constructed by the methods of statistical analysis and neural network have complex expressions and show little professional knowledge. However classification rules which classify samples directly based on the value of their attributes can be analyzed in the view of specialty. In this paper, the method of Chi-Merge based on entropy of information was applied to the discretization of continuous sample data. Then the problem of extracting optimal classification rules could be changed to a combinatorial optimization problem and solved through genetic algorithm. Thus, in this paper the problem of rule extracting was transformed to colony evolution. We also designed the fitness function of one individual. Once an optimal rule had been found, it was pruned. The pattern classifier was constructed by these pruned rules and a decision criterion. In the problem of the origin discrimination of olive oil, the classifier constructed by our method had been proved to have good performance with little number of classification rules and concise form and be appropriate to the classification of chemical patterns.

关 键 词：规则提取 规则修剪 遗传算法 化学模式分类器 连续属性离散化 判别准则 

分 类 号：TQ02[化学工程]